Institutionalized Defense Cooperation and Regime Security

Abstract

How does the institutionalization of external security cooperation affect government-military relations? This question has become increasingly salient amid a dramatic increase in bilateral defense cooperation agreements over the past few decades. I argue that DCAs help stabilize government-military relations and bolster regime security by mitigating a persistent credibility problem between signatory governments and their armed forces. By securing military-specific resources, DCAs limit the government’s ability to divert benefits elsewhere. The involvement of partner states further enhances credibility, as reneging on these bilateral agreements carries diplomatic and reputational costs. Because the military depends on the government to sustain these arrangements, the costs of plotting a coup increase: any extra-constitutional removal of the government threatens the institutional channels through which critical resources flow. A cross-sectional time-series analysis of 160 countries from 1980 to 2010 suggests that one or more DCAs exhibit significantly lower odds of coup attempts. The study highlights the role of international agreements in shaping domestic political stability. It also expands our understanding of states’ unique role in managing domestic civil-military relations through international brokerage and cooperation.

Introduction

International agreements influence domestic politics by providing local actors with access to resources, technology, or legitimacy, and by constraining their policy choices (see Boutton 2019; Pevehouse 2002; Simmons 2009). Some international agreements, such as formal alliances, require cooperation under specific circumstances, while others institutionalize relationships through regular exchanges of resources or information. Defense cooperation agreements (DCAs hereafter) are bilateral treaties that coordinate and institutionalize routine defense cooperation. Since the 1990s, they have become a predominant form of security cooperation regimes. Nearly a thousand DCAs are active globally, approximately 50 being signed each year (Kinne 2018, 801). Intriguingly, a significant portion of these agreements involves states that do not face common external military threats. For instance, Indonesia and Sweden, countries known for non-aligned foreign policies with no common security interests, signed a DCA in 2016. ¹

Understanding the effects of DCAs is crucial, as they differ significantly from formal defense alliances. DCAs do not commit countries to mutual defense in the event of an interstate conflict. Instead, they aim to facilitate resource sharing and capacity building. For instance, the DCA between Indonesia and Sweden includes defense-industrial collaboration, technology transfer, joint research and production, education and training-related exchanges, and military health services. ² The shifting global security environment with limited scope for new formal alliances boosts the demand for DCAs (Kinne and Bunte 2018). However, DCAs can engender a more profound effect on domestic politics by facilitating the transfer of resources and technology. Such transfers can help governments keep their military forces more satisfied and can rearrange the military’s incentives to intervene in politics.

I argue that DCAs enhance regime security by stabilizing government-military relations. Many governments are strained by a fundamental power-sharing dilemma rooted in a persistent credibility problem. The military cannot credibly commit to peacefully accepting rents without leveraging its coercive power, while the government cannot credibly commit to sustaining these rents over time without fear of military intervention (see Meng 2020; Meng and Paine 2022; Meng, Paine and Powell 2022; Paine 2021). This mutual distrust produces a fragile equilibrium in which the threat of intervention by either side undermines regime durability. DCAs help mitigate this problem by facilitating more enforceable and stable rent-sharing arrangements that externalize and institutionalize resource flows to the military.

DCAs make governments’ promises more credible in two ways. First, DCAs enable governments to secure non-fungible resources—such as advanced weapons, co-production facilities, specialized training pipelines, and high-level joint exercises—intended for their military. Because these benefits are defense-specific, governments face significant constraints on diverting them elsewhere. Second, DCAs strengthen the government’s commitments by involving external actors in the arrangement. Backing out of a DCA not only undermines relations with partner states but also signals unreliability to others. So, violating obligations risks diplomatic backlash, suspension of assistance, and reputational costs in broader international arenas. Mutual obligations also reinforce compliance in diverse partnerships. In asymmetric DCAs, stronger patrons incentivize cooperation because they also benefit, for instance, from basing rights, intelligence sharing, or joint industrial opportunities. In symmetric cases, both sides prefer compliance to maintain domestic credibility and to avoid diplomatic costs.

While both non-fungibility and external involvement enhance the government’s credibility, the military has strong incentives to maintain the status quo to ensure continued access to external benefits. Armed forces rely on a legitimate political authority to sustain these arrangements. Any extra-constitutional removal of the government disrupts the very institutional channel through which these resources flow and risks jeopardizing future access to critical military-specific benefits. So, the opportunity costs of plotting a coup rise considerably. After factoring in the risks of staging a coup and the costs of maintaining a military regime, the military is more likely to prefer civilian rule. Design features reinforce these dynamics. DCAs often integrate senior officers into joint steering committees and establish frameworks with long-term horizons. These features reduce reliance on a particular leader or a regime and ensure that the military can anticipate ongoing benefits across government turnovers. By securing military-specific resources, embedding external enforcement, and locking in institutional mechanisms, DCAs influence both government and military incentives and strengthen regime durability.

In Argentina, for instance, DCAs were associated with the Menem government’s success in minimizing military threats to civilian rule from the early 1990s onward. The government pursued a dual strategy of material inducements and military training. ³ DCAs with the United States, Spain, Italy, and others helped alleviate military grievances by providing enhanced prestige and international roles, such as participation in UN peacekeeping missions. These agreements included modernization initiatives and professional opportunities—such as sending high-ranking officers abroad for training and education—which incentivized the military to disengage from politics. This pattern is not limited to developing countries in Latin America, Africa, or Asia. For example, Greece, a member of the Organisation for Economic Co-operation and Development (OECD), also experienced the stabilizing effects of DCAs. Beginning in the late 1980s, the post-military government signed agreements with the U.S. and several European partners that supported modernization and provided access to advanced resources and technology. They helped the socialist government reduce the military’s political role and maintain civilian control. ⁴

I assess the relationship between DCAs and coup attempts using cross-sectional time-series data from 160 countries spanning 1980–2010. Baseline results are derived from mixed-effects logistic regression models estimated on both the full sample and a subsample of developing countries. Developing countries often have fragile government-military relations and weak institutional capacities, making them a more relevant sample. Across specifications—including pooled logit, random-effects models, extended covariate sets, and controls for time trends, regional clustering, and institutional variation—DCAs are associated with significantly lower odds of coup attempts.

To address concerns about selection bias—that countries signing DCAs may differ systematically from non-signatories—I first present descriptive evidence that these differences are minimal, particularly in terms of regime type or state capacity. I then implement an instrumental variables (IV) regression analysis using a lagged, distance-weighted measure of DCA diffusion among neighboring states to isolate plausibly exogenous variation. This IV model directly addresses threats of omitted variable bias and reverse causality, and the estimates reinforce the baseline findings. Finally, I conduct a difference-in-differences (DiD) analysis using group-time average treatment effects. This alternative identification strategy relies on pre- and post-treatment dynamics rather than exclusion restrictions. The results demonstrate that the relationship holds even when using timing and parallel-trend assumptions rather than an instrument.

This article makes several contributions. First, it identifies a mechanism by which governments address the credible commitment problem that drives many domestic political crises. Resources channeled through DCAs have greater built-in credibility because of their purely military use, enhancing the military’s stake in keeping the government in office. In addition, the government’s reneging costs increase, as commitments are institutionalized through international agreements. As a result, DCAs help disincentivize the military from intervening in politics.

Second, it offers insights into the role of governments as intermediaries between domestic actors and external powers in mitigating internal conflicts. In liberal theory, states’ policies emerge from aggregated preferences of key domestic groups, in which states function as a transmission or representation belt (Moravcsik 1997). In hierarchical relationships, a client state’s brokerage connects external powers to domestic groups, affecting the client state’s domestic political order (Lake 2013; Nexon 2009). I contend that by brokering agreements with other states, the government can help powerful domestic groups, particularly the military, gain access to substantial external resources, thereby mitigating domestic competition for resources.

Third, the paper shows how domestic and international orders are intertwined and, in part, co-constituted through cross-border security cooperation. International cooperation influences domestic actors’ preferences (Gourevitch 1978; Kastner and Rector 2003). Domestic politics also affect international outcomes, especially in alliance relations (Edry, Johnson and Leeds 2021; Kimball 2010; Simon and Gartzke 1996). This paper assesses how defense agreements, a fundamental building block of international order, help shape domestic order by stabilizing government-military relations and conditioning civil-military resource allocation.

Finally, this paper introduces a fresh analytic lens on the often-overlooked internal dimension of global security cooperation. Recasting DCAs in this way reorients the literature toward regime survival, foregrounding coup prevention, elite cohesion, and rent distribution as contingent yet crucial outcomes of security partnerships. In doing so, the paper offers insights into how institutional design and external third-party involvement can recalibrate intra-state bargaining and reshape civil-military relations.

Literature: Concepts and Contexts

Regime security refers to the protection of a government from being ousted by internal or external adversaries. Internally, military threats commonly manifest in coups d’état (Albrecht 2019; Paine 2022). Existing explanations emphasize regime type, political instability, economic performance, military grievances, and coup history (Finer 2002; Londregan and Poole 1990; O’Kane 1993; Sutter 1999). This study focuses on the role of international security agreements.

International Cooperation and Government-Military Relations

The literature that links international politics to civil-military relations remains limited, with most focusing on interstate conflict and external military interventions (Bueno de Mesquita, Siverson and Woller 1992; Zimmermann 2013). The external threat environment can heighten coup risk by empowering the military through increased investment in the defense sector (Feaver 1999; Gibler 2012; Houle 2016). It can also reinforce civilian control by fostering credible power-sharing arrangements (Desch 2008; McMahon and Slantchev 2015).

In the security domain, international agreements can provide material support to governments facing internal or external threats (David 1991). Domestic instability can be linked to the formation of external consultation and neutrality pacts (Edry, Johnson and Leeds 2021). The prospect of external intervention can enable authoritarian regimes to coup-proof—often at the cost of increased violence (Boutton 2019). So, the mechanism primarily involves signaling of external support in relation to domestic unrest or civil conflict.

DCAs go beyond signaling combat support. They coordinate resource transfers, build institutional capacity, and structure incentives of both the government and the military. These effects make DCAs more consequential for domestic politics and warrant a systematic study of their effects on regime security. Drawing inferences from formal alliances can be problematic, as they involve minimal resource transfers. This paper addresses this gap. First, it conceptualizes the institutionalization of everyday defense cooperation, distinguishing it from the episodic coordination typical of traditional military alliances. Second, it theorizes how such institutionalization facilitates credible and stable resource sharing between governments and their military forces. Third, it empirically tests whether DCAs enhance regime security using a robust cross-national time-series analysis.

Institutionalization of Everyday Defense Cooperation

The institutionalization of defense cooperation refers to the formal structures that enable regular coordination and governance of defense policies (Ashley Leeds and Anac 2005). It is organized through three key international regimes: (i) alliance treaties, (ii) DCAs, and (iii) informal strategic partnerships. Research centers on formal alliances and how peacetime institutional developments influence wartime joint operations, with provisions embedded in alliance treaties (Ashley Leeds and Anac 2005; Rapport and Rathbun 2021). Ashley Leeds and Anac (2005) classify institutionalization into three categories: joint preparation (e.g., integrated command, standardized policy, joint troop placement), moderate institutionalization (e.g., peacetime coordination, training, technology transfer, wartime subordination plans), and no institutionalization. Although about half of alliances require some institutionalization, fewer than 15 percent involve any peacetime contact between armed forces (Leeds et al. 2002). They do not include regular cooperation that operates beyond the shadow of a potential conflict or falls outside any formal alliance treaties.

DCAs are bilateral framework treaties that coordinate routine defense cooperation, including “(1) mutual consultation and defense policy coordination; (2) joint exercises, training, and education; (3) coordination in peacekeeping operations; (4) defense-related research and development; (5) defense industrial cooperation; (6) weapons procurement; and (7) security of classified information” (Kinne 2018, 803). Kinne (2020) categorize DCAs into two broad types: general, encompassing multiple areas, and sectoral, focusing on a single specific domain. ⁵ Although asymmetric forms existed earlier, their proliferation began in the 1980s and accelerated sharply after the early 1990s. Today, nearly one thousand DCAs are in force worldwide (Kinne 2018) (see Figure 1).

Figure 1.

Patterns of DCA proliferation (a) Over time (b) By categories.

DCAs are not confined to countries with historical or colonial ties, nor are they concentrated in any particular region, such as Europe. ⁶ For example, Indonesia and Brazil signed a DCA in 2011 despite lacking historical ties, common security threats, or prior defense relationships. These agreements typically entail long-term cooperation, averaging about a decade. ⁷ States may replace earlier agreements or consolidate several sectoral DCAs into a more comprehensive one. Most DCAs are symmetrically designed, making cooperation and obligations mutual and beneficial to both parties.

The institutionalization of defense cooperation through DCAs differs from other forms in three key dimensions (see Figure 2). First, DCAs “focus exclusively on cooperation” and therefore “impose a fundamentally different set of legal obligations than alliances do” (Kinne 2018, see appendix). They contain no provisions for mutual defense in the event of interstate conflict. For example, the 2002 Bangladesh-China agreement covered military training and defense production but made no commitments to wartime cooperation. ⁸ DCAs also differ from informal or loosely structured arrangements. ⁹ By design, they establish binding commitments to sustained, structured defense collaboration rather than ad hoc or symbolic cooperation.

Figure 2.

Variation in institutionalized defense cooperation.

Second, DCAs provide a broader scope for peacetime coordination and governance by facilitating a diverse set of non-combat, routine activities. Whereas alliances focus on capability aggregation and offer limited support for building domestic capacity (Morrow 1991), DCAs actively promote defense industries, military technology, and training infrastructure, reducing reliance on external security providers. For example, French DCAs with former African colonies strengthen partner states’ capacity to counter terrorism, combat illicit trafficking, and conduct peacekeeping operations. ¹⁰ The Sweden-South Korea DCA includes military-technical cooperation, defense intelligence sharing, and training. ¹¹ Such initiatives are sustained through institutional mechanisms—including joint committees, task forces, and working groups—that meet periodically to develop action plans and ensure compliance.

Finally, DCAs entail greater civil-military resource reallocation, allowing governments to secure substantial military assets and opportunities for skill development. Indonesia’s 2011 DCAs with China, Russia, India, Brazil, and Serbia facilitated the procurement and modernization of fighter jets, submarines, and missile systems. The 2005 Ukraine–Russia agreement permitted joint production of military hardware, whereas the Philippines’ 2006 DCA with Australia aimed to double military readiness through training exchanges. ¹² In many developing states, where the military poses a persistent threat to regime security (Kathman and Melin 2017), such agreements can help mitigate grievances by providing resources.

Theory: DCAs and Coups D’état

The institutionalization of everyday defense cooperation discourages military interventions in politics by facilitating a more credible and stable rent-sharing arrangement between the signatory government and its military. The argument does not claim that DCAs eliminate coups. Instead, by increasing the credibility of the government’s promises and incentivizing the military to preserve the status quo, DCAs can more effectively stabilize government-military relations, thereby reducing the average propensity to coup.

Civil-Military Rent-Sharing Dilemmas

Rent sharing between the government and the military is fraught with credible commitment problems. The first set of challenges arises when the military cannot credibly commit to accepting rents without resorting to force. The government faces a political dilemma in sharing rents due to the military’s dual role as both a threat to and a solution for regime security (McMahon and Slantchev 2015). Military elites leverage the government’s dependence on them to extract more rents, which undermines regime security by altering the domestic power balance. How the government responds to the military’s interests and influence, therefore, lies at the core of government-military relations. Commonly referred to as coup-proofing strategies, these responses include purging, counterbalancing, and appeasement. ¹³

Governments often prefer appeasement since purging and counterbalancing may backfire or reduce the military’s effectiveness against other threats. ¹⁴ Appeasement primarily involves transferring more benefits to the military (Bove and Nistico 2014; Collier and Hoeffler 2006; Meng 2019). While governments’ willingness to give in to military demands is understandable, whether this helps resolve unstable civil-military relations remains questionable. The incentives for both the government and the military to monopolize rents are pervasive. If the government anticipates future military intervention, it cannot credibly commit to honoring its promises or refraining from purging the military after buying time (Boutton 2019, 44). Without a credible commitment from the government, the military cannot commit credibly to refraining from using favorable shifts in power to alter the status quo.

A second set of challenges arises when the government fails to credibly promise continued rent sharing, incentivizing the military to intervene (see Bueno de Mesquita et al. 2005). The credibility issue may be more acute in regimes known for consolidating power, such as military dictatorships or personalist regimes. It can also emerge in other contexts in which military elites suspect that the regime might remove them from positions of power or restrict their access to resources and private benefits. ¹⁵

To discourage interventions, the government needs an arrangement that credibly promises the military adequate rents over time. The military generally prefers to respect civilian authority under a credible institutional framework, even if taking direct control can be tempting (Boix and Svolik 2013; Magaloni 2008; Roessler and Ohls 2018). The ex-ante credibility of military threats against a government that consistently keeps the military satisfied is low. ¹⁶ Moreover, intra-elite defections, public resentment, and external sanctions make military dictatorships costly to maintain (Clapham and Philip 2021). Thus, when generals are assured of adequate benefits, they are less likely to remove the government from office.

However, governments lack adequate mechanisms to make such promises credible. One option is appointing military elites to cabinet positions (Meng 2019), but that may not produce the intended resource transfers, and the military can exploit its access to the state apparatus to usurp power (Roessler 2016; Roessler and Ohls 2018). DCAs facilitate a more credible rent-sharing arrangement without exacerbating political dilemmas.

The Government’s Credible Commitment

DCAs help alleviate the credible commitment problems that arise in domestic rent-sharing arrangements. By offering the military access to substantial non-fungible external benefits, they limit the government’s ability to divert military resources elsewhere. They also increase the costs of government defection by adding third-party governments to the arrangements. I elaborate on these mechanisms with examples in the following subsections.

Non-fungible Resources

Most benefits secured through DCAs, including advanced weapons procurement, joint research and production facilities, specialized training pipelines, and high-level joint exercise opportunities, are designed for the military’s exclusive use and thus are not prone to budgetary or administrative diversion. ¹⁷ Because the military-specific resources cannot be repurposed outside the defense sector, the government’s rent-sharing promises become self-enforcing. It acts as a form of “lock-in” by design, requiring little institutional oversight. Once the deal is complete, even a predatory government has little room to shuffle resources elsewhere without significantly destroying their value.

Anecdotal evidence from around the world suggests that DCAs facilitate the sharing of non-fungible military benefits. For instance, Bangladesh’s DCA with India in 2017 was paired with a $500 million line of credit for Indian-made platforms and accompanied by regular training and capacity-building programs. ¹⁸ Because both the loan and the follow-up activities depended on the acquisition of specific Indian systems, reallocating the funds to civilian sectors would have breached the agreement’s technical specifications.

Vietnam presents a similar case. A $600 million Indian credit line funded the purchase of Indian patrol vessels and other maritime assets, directly enhancing Hanoi’s naval and air capabilities. ¹⁹ Given the hardware’s military-specific nature, any attempt to divert it would entail significant conversion costs and diplomatic backlash. In Indonesia, a 2015 cooperation agreement with Japan focused on licensing Japanese defense equipment and launching joint research and development, thereby entrenching supply chains and co-production lines that cannot be liquidated for other civilian benefits. ²⁰

The logic is also evident in Africa and Latin America. Nigeria’s 2023 South-South DCA with Brazil opened the door to joint production of military equipment, while a 2024 Nigeria–Egypt pact involved deeper defense-industrial collaboration—a domain inherently non-fungible because machinery, patents, and skill-transfer programs are valuable only within the defense ecosystem. ²¹ Many smaller states also leveraged this. The 2023 U.S.–Papua New Guinea Agreement pledges roughly $864 million over a decade for base upgrades, maritime surveillance assets, and troop training. ²² Likewise, a 2024 U.S.–Fiji pact funds patrol boats, medical logistics, and depot infrastructure for the military—all items whose utility outside the defense sector is minimal. ²³

So, the military-specific content of resources secured through DCAs can function as a commitment device. The more non-fungible these inflows are, the lower the government’s capacity to divert them. The military, aware of this structural constraint, needs fewer difficult-to-enforce institutional guarantees to believe that external benefits will indeed accrue to it.

Third-Party Involvement

The credibility of the government’s promises further increases because DCAs institutionalize rent-sharing through international agreements. This adds a mutually reinforcing layer of credibility, since partner states act as de facto third parties. They not only share resources and training for the military but also participate in bargaining over those opportunities. From the government’s perspective, violating obligations invites diplomatic costs, market retaliation, or a loss of access to advanced technology elsewhere (see LeVeck and Narang 2017; Sartori 2002). So, governments have strong incentives to maintain the promised resource pipeline to the armed forces. The evidence is compelling. Countries with DCAs have, on average, experienced a five-fold increase in arms imports, over a 200 percent increase in joint exercises, a 150 percent rise in training and exchange events, and more than a 100 percent increase in peacekeeping participation (Kinne 2018, 806).

Many case examples suggest that states that violate DCA obligations incur high international costs. South Korea’s 2019 decision to end its pact with Japan provoked sharp rebukes from Tokyo and Washington, undermining trilateral security coordination. ²⁴ Mali’s termination of its defense agreement with France in 2022 led to the withdrawal of French troops, the suspension of EU training, and the imposition of sanctions by the Economic Community of West African States. ²⁵ The expulsion of French special forces by Burkina Faso in 2023 collapsed its defense ties with France, ending intelligence-sharing and military assistance, and triggering a loss of trust among Western partners. ²⁶

Mutual benefits from DCAs, however, help ensure greater compliance. In asymmetric dyads, stronger patrons incentivize cooperation because they benefit through basing access, intelligence sharing, or joint industrial opportunities, and prefer to support rather than abandon faltering programs. For instance, the U.S. provided four additional pre-positioning sites, expanding cooperation with the Philippines, despite earlier domestic controversy. ²⁷ Similarly, the 2021 renewal helped sustain the British Army Training Unit, which trains about 1,000 members of the Kenyan Defense Force annually, while providing Nairobi with infrastructure and logistical support. ²⁸ When delays threatened Vietnam’s initial credit line, India not only refrained from imposing sanctions but also offered an additional $500 million in 2016. ²⁹

In more symmetric cases, both countries prefer compliance to make their commitments credible at home and avoid diplomatic or reputational costs. In the South Africa–Brazil A-Darter missile project, both co-financed R&D and certified the missile for their Gripen fleets, injecting additional funds to keep the program on track. ³⁰ Similarly, in the Argentina–Brazil nuclear safeguards regime, their joint agency conducts cross-inspections and files joint IAEA reports to avoid international penalties for breaching audited safeguards. ³¹ In both cases, enforcement is mutual because reputational risks keep cooperation credible and shared resources secure. Other cases from the Global South follow the same logic. In the KF-21 fighter program, South Korea renegotiated Indonesia’s delayed 20 percent cost share rather than cancel the project. ³² In the Indonesia–Turkey Kaplan MT tank program, firms signed a new production agreement in October 2024 to compensate for delays and maintain the partnership. ³³

Although unilateral withdrawals may create some uncertainty, past examples demonstrate two patterns. First, external patrons typically use partial suspensions to penalize human rights lapses or procurement irregularities. Washington’s September 2023 decision to withhold $85 million of Egypt’s Foreign Military Financing pressured Cairo without affecting hardware upgrades, co-production, or military training. ³⁴ Second, withdrawals generally follow extra-constitutional seizures of power by the military, precisely what DCAs aim to deter. France suspended all joint operations with Malian forces after the May 2021 coup, and the European Union froze cooperation with Niger within days of the July 2023 military takeover. ³⁵ In each case, the costs fell directly on the military that had ousted the civilian government.

The Military’s Status Quo Incentives

While both non-fungibility and third-party involvement help credibly secure external resources, the military has strong incentives to maintain the status quo to ensure its continued access to them. Armed forces depend on a legitimate authority to sustain these arrangements by fulfilling contractual obligations, matching donor commitments, and extending diplomatic cover for foreign personnel and contractors. Any extra-constitutional attempt to remove the government is costly for the military because it disrupts the very institutional channel through which these resources flow and risks jeopardizing access to critical military-specific benefits.

Many recent cases illustrate the high costs of disruption. After Egypt’s 2013 coup, the United States froze hundreds of millions in foreign military financing and halted deliveries of Apache helicopters and F-16 upgrades critical to the Egyptian Air Force. ³⁶ The pattern repeated in Niger in 2023, where the United States and the European Union halted all defense assistance within weeks of the coup, freezing non-fungible provisions such as UAV components and special forces training. ³⁷

Because these high-end, externally sourced facilities are hard to replace, the military faces a clear trade-off: preserve civilian rule and retain DCA-linked benefits, or seize power and risk losing an indispensable pipeline. Officers compare the short-term gains from a coup to the discounted value of future benefits that only a legitimate government can sustain. After factoring in the risks of staging a coup, the uncertainty of coup success, and the costs of maintaining a military regime, the opportunity costs of defection can significantly strengthen the military’s incentive to preserve the status quo. This incentive can persist even when DCA transfers strengthen the military’s coercive capacity. Appendix 1 formalizes the argument. Given that external benefits are large enough and the government complies, the military prefers to maintain civilian rule even if the domestic balance of power shifts in its favor.

Design features further reinforce this preference. A central mechanism is integrating military leaders into joint steering bodies, embedding the military within institutional structures that tie cooperation with the government’s continuity. For example, the annual dialogue under the India–Bangladesh DCA is co-chaired by India’s defense secretary and Bangladesh’s principal staff officer of the Armed Forces Division. ³⁸ A second feature is the time horizon. DCAs typically last a decade or more and are often renewed and expanded through procurement and industrial accords. As long as constitutionally legitimate authority is preserved, the military can anticipate ongoing benefits across government turnovers. By tying external rents to regime stability, DCAs turn the armed forces into stakeholders in the survival of civilian governments.

In addition, resources provided by DCAs, such as UN peacekeeping, distribute benefits widely across the military hierarchy, expanding the coalition within the military that favors political restraint (Sakib and Rahman 2023b). Troop deployments are rotation-based, enabling broad participation. Although Bangladesh contributes approximately 6,000 peacekeepers annually, more than 180,000 personnel have served over the past three decades. ³⁹ Allowances from these missions benefit individual officers while also funding procurement, modernization, and training infrastructure. The U.S.–Bangladesh DCA, for instance, helped upgrade the Bangladesh Institute of Peace Support Operation Training. ⁴⁰

In sum, DCAs deter both government opportunism and military adventurism by structuring incentives for both. When benefits are highly specialized, contingent on regime continuity, and institutionally embedded, these agreements stabilize civil-military relations, particularly in developing countries.

Hypothesis 1

States with one or more bilateral defense cooperation agreements are less likely to experience a coup d’état than countries without these agreements.

Research Design

Unit of Analysis and Sample

For analysis, I use data from 160 counties for the period 1980–2010. ⁴¹ The baseline analysis is at the country-year level, aligning with my monadic theoretical expectations. I also run additional dyad-year models for robustness checks, which control for both dyadic and country-level variables, including regime type, major power status, distance, etc.

Estimation involves both the full set of countries and only the developing countries. DCAs’ role can be more pronounced in the latter, as they typically face weaker institutional credibility and more fragile government-military relations (Kathman and Melin 2017). Moreover, resource constraints limit their policy options for managing domestic political actors. Given the definitional challenges of classifying developing nations, I exclude OECD members from the developing-country sample, as they are recognized as developed countries.

Variables and Data Sources

The dependent variable is a binary measure indicating whether at least one coup attempt occurs in a given year. A coup attempt indicates “illegal and overt attempts by elites within the state apparatus to unseat the sitting executive” (Powell and Thyne 2011, 252). The military often plays a significant role in coups initiated by non-military elites. Powell and Thyne (2011) do not restrict their definition to military-led coups, unlike Finer (2002), to address the bias that military involvement is frequently hidden unless the attempt succeeds. They exclude civilian and mercenary movements, focusing instead on plots involving state elites. My theory also applies when non-military elites initiate plots with overt or covert military support. Additionally, I do not restrict the analysis to successful coups, as success can depend on factors beyond the scope of this study.

The primary explanatory variable is bilateral DCAs. I draw on the dataset by Kinne (2018), which defines DCAs as framework treaties that institutionalize routine defense cooperation between states. These agreements cover activities such as policy coordination, arms procurement, training, exercises, peacekeeping, and officer exchanges. I measure DCA involvement using two indicators: (1) the maximum number of DCAs in force in a year and (2) a binary indicator for whether at least one DCA is active. I include both general and sectoral DCAs, as each confers tangible benefits to the armed forces.

The base models include several covariates. Defense alliance membership is a binary variable indicating whether a country is part of a formal defense alliance (Leeds et al. 2002). Alliances represent another form of institutionalized security cooperation and can generate both network effects and norm diffusion, encouraging states to enter additional DCAs (Sommerer and Tallberg 2019). Alliances may also reshape domestic civil-military relations by imposing institutional constraints on armed forces (Boutton 2019).

To account for prior instability, I include past coups, using a weighted average of total attempted coups per year prior to the observation year (Powell and Thyne 2011). A history of military intervention increases the risk of future coups (Belkin and Schofer 2003; Roessler 2011) and may also shape a country’s propensity to enter DCAs as a regime-stabilization mechanism.

Democracy measures regime type using Polity IV scores ranging from −10 (full autocracy) to 10 (full democracy) (Marshall et al. 2002). Autocratic regimes are generally more vulnerable to coups (Bjørnskov and Rode 2019), while democratic states tend to participate in international agreements more (Lai and Reiter 2000; Mansfield, Milner and Rosendorff 2002). Robustness checks also use a dichotomous measure of democracy (Boix, Miller and Rosato 2013). To capture state capacity, I include log GDP per capita, measuring economic development and military capability. For robustness, I also include GDP growth, measured as the annual percentage increase in per capita income. Both variables are drawn from Graham and Tucker (2019).

In models with an extended set of covariates, I incorporate log-transformed direct measures of military capability, including annual military expenditure (in millions, constant 2011 USD) and military personnel (Singer et al. 1972). These variables capture the armed forces’ coercive capacity—which may increase their domestic political leverage—while also indicating that stronger states are more likely to engage in international security cooperation and less likely to experience coups.

The extended models include four covariates capturing external and internal military threats, which shape security cooperation and civil–military relations. External threat is measured using the aggregated Index of National Capabilities (CINC) scores of likely adversaries, defined as non-allied contiguous states with divergent foreign policy preferences or major powers with opposing alignments (Edry, Johnson and Leeds 2021; Leeds and Savun 2007). I also include an indicator for interstate conflict, coded 1 if the country engaged in a fatal militarized interstate dispute that year (Palmer et al. 2020).

Domestically, I include a measure of excluded ethnic groups with large trans-border ethnic kin (Vogt et al. 2015), which can drive external security cooperation due to threats to sovereignty and internal stability (Edry, Johnson and Leeds 2021). I also control for civil war, coded 1 if a conflict occurred in the previous year (Gleditsch et al. 2002), as post-conflict periods heighten vulnerability to military takeovers and external intervention. Summary statistics are reported in Tables A1 and A2 in Appendix 2.

For the descriptive statistics, I present the distribution of countries with at least one attempted coup (graph a) and with at least one DCA (graph b) in Figure 3. While the number of countries experiencing coups has generally declined, the data show persistent year-to-year fluctuations and occasional spikes. In contrast, DCAs rapidly proliferated: roughly 140 countries had signed at least one DCA by 2010. The figure also disaggregates DCAs’ prevalence by regime type (graph c) and development status (graph d). The spread of DCAs is not confined to democracies or developed states—both of which are typically less susceptible to military intervention in politics. By 2010, around 60 autocracies and 80 democracies had at least one DCA in effect. Although developed countries initially dominated DCA participation, this changed during the 1990s. By 2010, over 100 developing countries were integrated into the global DCA network.

Figure 3.

Descriptive statistics (a) Coup attempts by year (b) DCAs by year (c) DCAs and regime type (d) DCAs and development status.

Additional graphs (see Figure A3 in Appendix 6) indicate that only 43 DCAs have been terminated (graph (a)), whereas nearly 800 remain active. ⁴² Among the terminated DCAs, the majority lasted at least 10 years (graph (b)). A similar duration pattern emerges when both terminated, and active agreements are considered together (graph (c), suggesting that most DCAs — regardless of their eventual status — tend to be long-lasting. ⁴³

Baseline Results

Mixed Effects Logistic Regression Analysis

The baseline analysis uses mixed-effects (ME) logit models to assess the hypotheses involving a binary outcome variable. These models account for country-specific unobserved heterogeneity without being subject to the limitations of random-effects (RE) logit (i.e., biased estimates if unobserved effects are correlated with regressors) or fixed-effects (FE) logit (i.e., the incidental parameter problem). ⁴⁴ The specification for the ME logit model is:

$\mathrm{Pr}(Y_{i,t}=1)={\text{logit}}^{-1}(\alpha +X_{i,t}^{\prime }\beta +{\lambda }_i+{\epsilon }_{i,t})$

(1)

In this equation, α denotes the global intercept, and β captures the effects of observed covariates X_i,t, which are treated as fixed effects. The term λ _i represents unit-specific random intercepts, assumed to follow a normal distribution, ${\lambda }_i\sim \mathcal{N}(0,{\sigma }_{\lambda }^2)$ , capturing unobserved country-level heterogeneity. ɛ_i,t is the residual error term. This formulation allows for the accounting of both observed and unobserved heterogeneity across units. The logit link ensures that predicted probabilities remain bounded between 0 and 1.

Table 1 reports four mixed-effects models. Columns 1–2 use the full sample, and columns 3–4 restrict to developing countries. These results support Hypothesis 1, which posits that DCAs reduce military intervention in politics. Substantively, more DCAs are associated with a lower likelihood of coup attempts, and countries with at least one DCA face a lower coup risk than those without. The pattern holds in developing countries (Models 3–4), where effects are slightly larger, indicating the findings are not driven by developed countries.

Table 1.

Mixed effects logit models

	Dependent variable: Coup attempt
DCA (count)	−0.550***		−0.617***
	(0.167)		(0.188)
DCA (dummy)		−0.734***		−0.763***
		(0.214)		(0.216)
Defense alliance	0.469**	0.477**	0.437*	0.455*
	(0.238)	(0.242)	(0.236)	(0.241)
Past coup	2.196***	2.204***	2.225***	2.217***
	(0.696)	(0.703)	(0.691)	(0.699)
Democracy (Polity IV)	−0.0434***	−0.0494***	−0.0386**	−0.0437***
	(0.0161)	(0.0163)	(0.0163)	(0.0164)
GDP per capita (logged)	−0.391***	−0.491***	−0.360***	−0.448***
	(0.0988)	(0.103)	(0.104)	(0.109)
Constant	−1.064	−0.384	−1.219	−0.619
	(0.683)	(0.712)	(0.710)	(0.748)
Observations	4524	4524	3719	3719
Developing country sample			✓	✓
Fixed effects	✓	✓	✓	✓
Random intercepts	✓	✓	✓	✓

*p < 0.1.

**p < 0.05.

***p < 0.01.

Figure 4 shows that coup odds decline by more than half in countries with one or more DCAs, in both the full sample (panels A–B) and developing-country subsample (panels C–D). Additional DCAs further reduce coup risk, but the marginal effect diminishes and becomes non-significant once a country reaches about 8–10 DCAs (Figure A4 in Appendix 6).

Figure 4.

Odds ratios for coup attempt.

Robustness Checks

The results are robust to an extended set of controls, one-year lags, and alternative measures of GDP and regime type (Tables A3-A5 in Appendix 3). They also hold with alternative time controls, including a Cold War dummy, linear year trends, and 5- and 10-year interval fixed effects (Tables A6 and A7 in Appendix 3). Findings are similar in random-effects and pooled logit models, with and without cubic polynomials of years since the last coup attempt (Carter and Signorino 2010) (Tables A8-A10 in Appendix 3).

Adding regime-type and regional fixed effects (Cheibub, Gandhi and Vreeland 2010) does not change the results (Table A11 in Appendix 3). Coup risks are not randomly distributed across regimes—military regimes being more prone to coups—and regional clustering is evident in Sub-Saharan Africa and Latin America. The results remain significant and substantively similar under these stricter specifications. The findings also hold in dyad-year models, including specifications that exclude the United States and the UN Security Council P5 from the country-i sample (Tables A13-A15 in Appendix 4).

The analysis also speaks to alternative mechanisms. If coups are deterred mainly by fear that external patrons (e.g., the United States) would oppose regime change and suspend cooperation, defense alliances—not DCAs—should matter more (Boutton 2019). Instead, alliances are either positive or null across models, offering no consistent support for an intervention-threat account.

A second mechanism is socialization, as DCAs often include training and joint exercises that may strengthen professional norms (Kinne 2018; Ruby and Gibler 2010). Yet these provisions are typically bundled with material aid and treated as tangible benefits by recipient military forces (Martinez Machain 2021). If socialization were central, DCA effects should be stronger in democracies, but DCA × democracy interactions are not significant in the robustness models (Table A12 in Appendix 3). Dyad-year models provide limited evidence that DCA ties between two democracies are slightly stronger than those involving at least one democracy (Table A14 in Appendix 4).

Addressing Selection Biases

Descriptive Patterns

Countries that sign DCAs may differ systematically from those that do not, creating selection bias. States with lower coup risk may be more likely to sign DCAs, while potential partners may avoid high-coup-risk states. Regime type shapes both the government’s incentives to cooperate internationally and the military’s incentives to intervene (Bjørnskov and Rode 2019). If autocracies are more coup-prone and less likely to sign DCAs than democracies, estimates will be biased. Weaker state capacity also increases the risk of coups. So, if low-capacity states are underrepresented among signatories, estimates of DCA effects will be biased too.

Descriptive trends, however, suggest that this selection is not likely. As Figure 3 presented earlier shows, autocracies and developing countries have increasingly signed DCAs, often at rates comparable to their democratic and developed counterparts. Indeed, states that have experienced periodic military interventions—including Pakistan, Bangladesh, Egypt, Indonesia, Argentina, and so on—have entered into multiple DCAs. ⁴⁵ Many Sub-Saharan African states have also signed DCAs with one or more partners (Kinne 2018, p 807), suggesting that states vulnerable to military influence are not excluded from the diffusion of DCAs.

Instrumental Variable Regression

While I include many covariates to account for observable confounding variables and conduct robustness checks, they cannot fully address concerns about reverse causality. To mitigate this, I estimate an instrumental variables (IV) regression using a diffusion-based instrument that captures the weighted, lagged trends in DCA proliferation among neighboring states, following the logic of spatial diffusion (Elkins, Guzman and Simmons 2006). For each country i, I calculate the total number of DCAs signed by its neighboring countries j with their partner countries p, weighting the counts by the inverse of the geographic distance between country i and each neighbor j. I then average these weighted values by country-year and lag the resulting measure by a year to ensure temporal precedence relative to country i’s DCA signing behavior. This instrument captures spatial-temporal diffusion effects while remaining plausibly exogenous to country i’s internal coup risk or political dynamics.

This instrument leverages exogenous variation in terms of neighboring countries’ propensity to sign DCAs, since expectations of a coup in country i are unlikely to drive DCAs in those states. Temporal lagging reduces the risk of simultaneous determination between the endogenous regressor and the instrument. ⁴⁶ The instrument’s influence on regime security is expected to operate through alternative forms of security cooperation or via external threats that shape those partnerships. So, the models control for formal alliances and external threats. The instrument also meets the relevance criterion, as demonstrated by strong first-stage F-statistics indicating a robust correlation with the DCA variables.

Table 2 presents the second-stage estimates from the two-stage least squares models with first-stage results reported in Table A16 in Appendix 5. The instrument is highly statistically significant and positively correlated with the treatment variable. The instrumented coefficients of the effect of DCAs on coup attempts remain statistically significant and in the expected negative direction. The results are robust to an expanded set of covariates (see Tables A17 and A18 in Appendix 5). The magnitude of the negative coefficients is larger in the subsample of developing countries, consistent with earlier results that suggest a more stabilizing effect of DCAs in these countries.

Table 2.

Two-stage least square regression models

	Dependent variable: Coup attempt
DCA (count) instrumented	−0.00346**		−0.00572**
	(0.00155)		(0.00268)
DCA (dummy)instrumented		−0.0315**		−0.0325**
		(0.0135)		(0.0145)
Defense alliance	0.0107	0.0107	0.00672	0.0115
	(0.00596)	(0.00570)	(0.00757)	(0.00682)
Past coup	0.226***	0.224***	0.222***	0.224***
	(0.0446)	(0.0442)	(0.0455)	(0.0449)
Democracy (Polity IV)	−0.000223	−0.000743**	−0.000106	−0.000956**
	(0.000453)	(0.000388)	(0.000598)	(0.000435)
GDP per capita (logged)	−0.00458**	−0.00613***	−0.00501	−0.00858***
	(0.00230)	(0.00184)	(0.00317)	(0.00219)
Constant	0.0613***	0.0814***	0.0705***	0.0980***
	(0.0161)	(0.0139)	(0.0201)	(0.0175)
Observations	4386	4386	3605	3605
Developing country sample			✓	✓
First-stage R2	0.2281	0.2546	0.1693	0.2561
First-stage F statistics	16.25	67.96	9.81	62.17

The first stage results are reported in Appendix 5 in Table A16.

*p < 0.1,

**p < 0.05,

***p < 0.01.

Group-Time Average Treatment Effects

A DiD framework is well-suited for examining trends in coups before and after shifts in defense cooperation. Two key identification assumptions are required for the DiD estimation (Ashworth, Berry and De Mesquita 2021). First, after controlling for time-varying confounders, the treated and control groups do not exhibit systematic differences in the outcome prior to treatment. Second, no other factors change contemporaneously with the treatment in a way that differentially affects the two groups. I estimate a GTATE model—the staggered adoption of the DiD framework—to strengthen the credibility of the baseline results.

GTATE models estimate the average treatment effect for each group and time, with the timing of initial treatment defining the group (Callaway and Sant’Anna 2021). They offer a better alternative to two-way fixed effects when treatment timing varies across groups and over time (De Chaisemartin and d’Haultfoeuille 2023; Imai and Kim 2021; Kropko and Kubinec 2020; Roth et al. 2023). In this framework, each dyad is assigned a treatment-onset year, and effects are computed for subsequent periods by comparing treated units with “not-yet-treated” or “never-treated” units at the same time point. This structure accounts for treatment-effect heterogeneity and avoids the biases inherent in TWFE models.

To assess the validity of the identification assumptions, I examine whether pre-treatment trends of treated and control units follow similar trajectories (Callaway and Sant’Anna 2021; Sun and Abraham 2021). I estimate trends for 30 years prior to the first year of treatment. Figure 5 presents these trends for four models: two for the full sample and two for the developing-country sample. The results show no meaningful differences in outcome trends in the pretreatment period, even after controlling for covariates. The graphs support the parallel trends assumption, enhancing the credibility of the GTATE estimates and suggesting that post-treatment differences reflect changes in defense cooperation rather than pre-existing trends.

Figure 5.

Parallel trends.

The GTATE estimates are aggregated for the Average Treatment Effect on the Treated (ATT), analogous to the canonical DiD estimator. Across all four models, the ATT results show that states with a DCA in place are significantly less likely to experience a coup (see Figure 6), consistent with previous results and reinforcing the core claim that institutionalized defense cooperation reduces the likelihood of coups.

Figure 6.

Average treatment effects on the treated.

Discussion

Recent research on rent-sharing institutions in civil-military relations affirms DCA’s stabilizing effects (Meng 2019; Meng, Paine and Powell 2022; Sakib and Rahman 2023a, 2023b). Studies showing that military cooperation helps countries reduce their propensity to coups also offer support (Atkinson 2014; Ruby and Gibler 2010). A few studies that claim the opposite effects either focus on a single factor, such as military training, or limit their scope to highly coup-prone, autocratic countries only (Böhmelt, Escribà-Folch and Pilster 2019; Joyce 2022; Savage and Caverley 2017). The empirical evidence on military training is mixed (Savage 2021; Tecott, Urben and Grewal 2022). A recent study with more disaggregated data shows that the adverse effects of military training are “limited to a single U.S. program representing a small fraction of overall U.S. training activities” (McLauchlin, Seymour and Martel 2022, 282).

Prior studies also support my findings related to alternative explanations. Defense alliances worsen government-military relations (Boutton 2019). The lack of socialization effect is supported by many recent studies (Savage and Caverley 2017; Taylor 2014). My findings show that DCAs, as a broader category that provides numerous institutionalized external benefits for the military, help stabilize civil-military relations on balance.

Conclusion

How security cooperation affects government-military relations remains understudied. DCAs, bilateral platforms that facilitate the military’s access to international resources, are particularly relevant to this salient question. These agreements have emerged as pivotal international regimes that help signatory states to institutionalize their everyday defense cooperation. Varying substantially from defense alliances in legal obligations, peacetime governance, and civil-military resource allocation, they help manage troubled government-military relations and change the military’s incentives to intervene directly in politics. Many provisions under DCA arrangements, including arms procurement, research, training, and industrial development, and peacekeeping appointment opportunities, also help governments address military grievances. Securing resources from external partner states through international commitments and the military-specific use of those resources offer a more credible solution to commitment problems in government-military competition for rents.

The empirical findings show that states with DCAs have lower odds of experiencing military interventions. The findings apply to a world sample and, more specifically, to developing countries. DCAs’ effects are not conditional on the great powers’ support for client states. Effects are generalizable worldwide. Countries with more DCAs experience greater stabilizing effects on their domestic politics. So, signing more DCAs adds some value. However, more traditional forms of military cooperation, such as formal alliances, do not consistently affect coups. Moreover, democratic countries do not always derive greater benefits from DCAs, suggesting that DCAs’ socialization effects are minimal or, at best, limited to democratic dyads.

The study offers insights into how international factors affect domestic politics, with potential implications for regime durability and democratization. International agreements facilitate civil-military resource reallocation, thus rearranging the military’s incentives to intervene in politics. DCAs, as international obligations involving partner states, enhance the government’s credibility with its armed forces. Overall, the paper expands our understanding of the role of international agreements in domestic politics, identifies states’ unique role in managing domestic civil-military relations through international brokerage, and shows how domestic and international orders are co-constituted through cross-border security cooperation.

Supplemental Material

Supplemental Material - Institutionalized Defense Cooperation and Regime Security

Supplemental Material for Institutionalized Defense Cooperation and Regime Security by Md Muhibbur Rahman in Journal of Conflict Resolution.

Supplemental Material

Supplemental Material - Institutionalized Defense Cooperation and Regime Security

Supplemental Material for Institutionalized Defense Cooperation and Regime Security by Md Muhibbur Rahman in Journal of Conflict Resolution.